[SPARK-18940] Percentile and approximate percentile support for frequency distribution table

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Incomplete
• Affects Version/s: 2.0.2
• Component/s: SQL
• Labels: bulk-closed
• Assignee: Unassigned
• Reporter: gagan taneja
• Shepherd: Herman van Hövell
• Created: 20/Dec/16 08:11
• Updated: 21/May/19 04:17
• Resolved: 21/May/19 04:17

Description

I have a frequency distribution table with following entries

Age,    No of person 
21, 10
22, 15
23, 18 
..
..
30, 14

Moreover it is common to have data in frequency distribution format to further calculate Percentile, Median. With current implementation
It would be very difficult and complex to find the percentile.
Therefore i am proposing enhancement to current Percentile and Approx Percentile implementation to take frequency distribution column into consideration
Current Percentile definition

percentile(col, array(percentage1 [, percentage2]...))
case class Percentile(
  child: Expression,
  percentageExpression: Expression,
  mutableAggBufferOffset: Int = 0,
  inputAggBufferOffset: Int = 0) {
   def this(child: Expression, percentageExpression: Expression) = {
    this(child, percentageExpression, 0, 0)
  }
}

Proposed changes

percentile(col, [frequency], array(percentage1 [, percentage2]...))
case class Percentile(
  child: Expression,
  frequency : Expression,
  percentageExpression: Expression,
  mutableAggBufferOffset: Int = 0,
  inputAggBufferOffset: Int = 0) {
   def this(child: Expression, percentageExpression: Expression) = {
    this(child, Literal(1L), percentageExpression, 0, 0)
  }
  def this(child: Expression, frequency : Expression, percentageExpression: Expression) = {
    this(child, frequency, percentageExpression, 0, 0)
  }
}

Although this definition will differ from hive implementation, it will be useful functionality to many spark user.
Moreover the changes are local to only Percentile and ApproxPercentile implementation